A pause rule for Japanese text-to-speech conversion technique is proposed, which can determine natural pause locations. In order to insert several pauses at appropriate bunsetsu boundaries (which resemble "phrase" boundaries in English), the probabilities (pause insertion probabilities) that words are followed or preceded by pauses are used. The pause insertion probabilities are obtained by statistically analyzing a large number of sentence utterances. It was found that the probabilities differ from each other, according to the parts of speech for the words adjacent to the pauses. By the rule, adequate pauses are inserted at the bunsetsu boundaries whose pause insertion probabilities are high. An evaluation experiment for the rule was carried out, using 200 sentences. The result indicates that the pause locations, determined by the rule, are as natural, in 93% of the sentences, as those determined by humans. The rule is adopted by a Japanese text-to-speech conversion system.
Cite as: Iwata, K., Mitome, Y., Watanabe, T. (1990) Pause rule for Japanese text-to-speech conversion using pause insertion probability. Proc. First International Conference on Spoken Language Processing (ICSLP 1990), 837-840, doi: 10.21437/ICSLP.1990-107
@inproceedings{iwata90b_icslp,
author={Kazuhiko Iwata and Yukio Mitome and Takao Watanabe},
title={{Pause rule for Japanese text-to-speech conversion using pause insertion probability}},
year=1990,
booktitle={Proc. First International Conference on Spoken Language Processing (ICSLP 1990)},
pages={837--840},
doi={10.21437/ICSLP.1990-107}
}